In the format provided by the authors and unedited.

1

Supplementary Information for

Theory-driven design of high-valence metal sites for water

oxidation confirmed using in situ soft X-ray absorption

Xueli Zheng†, Bo Zhang†*, Phil De Luna†, Yufeng Liang, Riccardo Comin, Oleksandr

Voznyy, Lili Han, F. Pelayo García de Arquer, Min Liu, Cao Thang Dinh, Tom

Regier, James Dynes, Sisi He, Huolin L. Xin, Huisheng Peng, David Prendergast,

Xiwen Du and Edward H. Sargent*

†These authors contributed equally to this work

*Correspondence and requests for materials should be addressed to Edward H.

Sargent (ted.sargent@utoronto.ca) (E.H.S) and (boo.zhang@utoronto.ca) (B.Z.)

Contents:

Theoretical Calculations …………………………………………………………2

Materials and Methods …………………………………………………………...9

Supplementary Figures …………………………………………………………..16

Supplementary Tables ……………………………………………………………45

Structures of Models Used ……………………………………………………….50

References ………………………………………………………………………..51

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

SUPPLEMENTARY INFORMATIONDOI: 10.1038/NCHEM.2886

NATURE CHEMISTRY | www.nature.com/naturechemistry 1

2

Theoretical Calculations.

Energetics Calculations

The initial structure of Ni(OH)2 was taken from literature reports36, while

β-NiOOH was chosen with staggered intercalating protons in light of recent work23

that found that a stable low-spin d7 Ni3+ configuration was consistent with

experimental crystal structures. For NiO2, the initial structure was derived from the

crystal structure of Ni(OH)2 with all protons deleted from the unit cell. Full geometry

and unit cell optimizations were performed on the base Ni(OH)2, β-NiOOH, and NiO2

structures before adding the effect of metal and phosphorus doping. The Gibbs free

energy of formation, ΔG°f, was calculated using the framework of Persson et al.17 that

has previously been used to study Co oxides and (Ni,Fe)OOH for OER energetics13,

24. Bulk energetics of metal and phosphorous doped Ni-oxides were calculated using

density functional theory (DFT) with the Hubbard-U framework (DFT+U) to account

for the strongly localized d-electrons for Ni, Fe, and Co37. The Hubbard-U correction

terms were fixed at Uhub(Ni) = 6.45 eV, Uhub(Co) = 3.32 eV, and Uhub(Fe) = 5.3 eV as

obtained via linear response theory38. These values were taken from previous studies

on Ni,Fe(OOH) and Co oxides13, 22, 24. Furthermore, it has been shown that varying

Uhub(Ni) values between 2.5 eV to 7 eV showed no qualitative change in optimized

lattice parameters22. The Perdew-Burke-Ernzerhof (PBE) exchange-correlation

functional39 and Blochl’s all-electron, frozen-core, projector augmented (PAW)

method40,41 with a plane-wave basis set cutoff value of 750 eV were used with the

Vienna Ab-initio Simulation Package (VASP)42. PAW potentials were in effect for

the 4s/3d electrons of Ni, Co, Fe, and the 2s/2p electrons of O, while all other core

electrons were kept frozen. The total energy was considered converged once the

energy difference reached below 1 meV/atom. Γ-point-centered Monkhorst−Pack44

was used and only the gamma point was sampled for the k-points for optimizations.

Bulk structures were 6x2x2 supercells and were converged below a force threshold of

0.01 eV/Å with Gaussian smearing and a smearing width of 0.05 eV. For density of

states, a 3x9x9 Γ-point-centered Monkhorst−Pack k-point mesh was used. The

3

computational hydrogen electrode (CHE) was used to calculate the change in free

energies (ΔG) for the oxidation of Ni2+ Ni3+ and Ni3+

Ni4+.

Optimized Ni(OH)2 lattice constants match well with experiment39 and β-NiOOH

lattice constants agreed with the previous DFT study with intercalated protons22. Table

S1 shows lattice constants for all optimized structures. First, binary mixtures were

modeled by random substitution of two Ni atoms in 6x2x2 lattice structures of Ni(OH)2,

β-NiOOH, and NiO2 with Co and Fe. Ternary mixtures were made through random

substitution of four Ni atoms, two for Fe and two for Co. Next, NiO2Hx, NiCoO2Hx, and

NiCoFeO2Hx structures were doped with phosphorus through substitution of one Ni

atom for a P in order to mimic the incorporation of a phosphate in the bulk structure. All

doped structures were also allowed to be fully optimized.

Spin polarized calculations were performed on all structures. Initial Ni(OH)2,

NiOOH, and NiO2 starting structures were chosen with anti-ferromagnetic

arrangement which was previously shown to be preferred compared to the

ferromagnetic or parallel spin configurations22. The magnetic moments for each

nickel ion in the optimized low spin structures was found to be approximately ± 1.8

μB, ± 1.2 μB, and ± 0 μB for Ni(OH)2, NiOOH, and NiO2 respectively. The magnetic

moment for NiOOH is similar to previously published low-spin magnetic moment of

± 1.5 μB from ab-initio calculations22.

Since the experimental catalyst is mainly amorphous, we are not able to provide a

structural model at long range using DFT. Thus, we have decided to use a model system,

with the assumption of the preservation of local chemistry, to elucidate trends with

respect to doping of other elements. γ-phase is significantly more complex to model

computationally due to the larger number of possible intercalating species and different

stacking orientations of the NiOx sheets. To properly and thoroughly model all possible

configurations of doped γ-NiOx would be prohibitively time intensive. Furthermore,

there has been extensive work on the structural and electronic features of β-Ni(OH)2

and β-NiOOH on which we could lean to ensure accuracy of our models42, 43. Therefore,

we believe that the use of the β-phase is justified in this study.

4

The balanced chemical reaction was calculated at pH=0 and was referenced against

the standard hydrogen electrode. Therefore, for a given nickel oxyhydroxide the

oxidation reaction is (NiOxHy NiOxHy-1 + H+ + e-) and the reduction reaction is (2(H+

+ e-) H2). The Gibbs free energy of the H+ is accounted for in the chemical potential

of a proton as calculated in the framework provided by Persson et al14. Simply, we

begin with oxygen and hydrogen in equilibrium with water in an aqueous environment,

½ O2(g) + H2(g) H2O(l). Next the chemical potential of water is written as a function

of the oxygen and hydrogen Gibbs free energies, μH2O = ΔGH2O = GH2O – GH2 – ½ GO2.

The reference Gibbs free energy for hydrogen is now GH = ½ [GH2O - ΔGH2O – ½ GO2].

The chemical potential of hydrogen in the standard state can now be written as μH = ½

[ EH2Odft – TSH2O – μO – μH2Oref] where EH2Odft = -14.7 eV, T = 298 K, SH2O = 7.24 x

10-4 eV/K, μO = -4.57 eV, μH2Oref = -2.46 eV and μH = -3.73 eV. This value is

consistent with our calculations because we have used the same DFT package (VASP)

implemented with the identical PBE GGA functional and the same convergence cutoffs

of 5 meV/atom.

The Gibbs energy of formation (ΔG°f) for a nickel oxyhydroxide is outlined below

as ΔG°f (NiOxHy) = ENiOxHy – ENi – xEO – yEH where ENi is -7.6 eV (calculated from

bulk nickel), EO is -4.57 eV (calculated from water), and EH is -3.73 eV (as calculated

from water described above). The Gibbs energies of formation are listed in table S2.

The Gibbs energy of oxidation from Ni2+ to Ni3+ (ΔG(Ni2+ Ni3+)) is calculated as

ΔG(Ni2+ Ni3+)= ENiOOH – ENi(OH)2 – μH where ENiOOH is the energy of NiOOH as

calculated from DFT, ENi(OH)2 is the energy of Ni(OH)2 as calculated from DFT, and μH

is the chemical potential of hydrogen. The entropy was not calculated for the bulk

catalyst since the energy contributions due to the vibrations of the bulk are very small in

comparison to water or hydrogen35.

We believe this value is consistent with our calculations because we have used the

same DFT package (VASP) implemented with the PBE GGA functional and the same

convergence cutoffs of 5 meV/atom. Supplementary Table 2 shows the DFT energies

per formula unit and the formation free energies for all bulk structures. It was found that

5

NiO2 has a positive free energy of formation which shows that this structure is not

preferred. This matches experiment as the NiO2 phase is not seen in the Ni Pourbaix

diagram. However, the doping of Co, Fe, and P shows that this phase has a negative

ΔG°f..

The change in Gibbs energy for the oxidation of Ni was calculated using the

following equations:

ΔG(Ni2+ Ni3+)= ΔGf(NiOOH)-[ ΔfG(Ni(OH)2+µH] (1)

ΔG(Ni3+ Ni4+)= ΔGf(NiO2)-[ ΔfG(NiOOH+µH] (2)

where µH is -3.73eV, and ΔGf(NiOx) is the formation energy of the various structures

as listed in Supplementary Table 2. Using experimental formation energy values for

Ni(OH)2 and NiOOH the ΔG(Ni2+ Ni3+) is -2.39 eV which is in agreement with the

calculated ΔG(Ni2+ Ni3+) of -2.24 eV46.

To account for the thermodynamics of the electrochemical environment we have

calculated the ΔG(Ni2+ Ni3+) and ΔG(Ni3+ Ni4+) as a function of pH. The slope of the line

is given by E0 = ΔG°f (Ni3+) – ΔG°f (Ni2+) – 0.05916*pH for the Ni2+ to Ni3+ oxidation

and E0 = ΔG°f (Ni4+) – ΔG°f (Ni3+) – 0.05916*pH for the Ni3+ to Ni4+ oxidation. The

plot (Supplementary Fig. 25) shows that at pH=7 the anodic potential window for Ni3+

to exist is much larger (0.2V – 1.1V) for the doped NiCoFeP catalyst than for Ni alone

(1.07V – 1.29 V). Notice that at the applied potentials of 1.4V and above the Ni4+

species should be stable.

The β-NiOx was chosen as a model system to study the effect of intimate elemental

doping. The experimental catalyst is mainly amorphous without a crystalline LDH

phase and thus it is not possible to accurately model such a system using DFT. We have

chosen a model system assuming the preservation of local chemistry to study the effect

of doping with other elements. This model system strategy has been previously

employed to study an amorphous FeCoW oxyhydroxide OER catalyst16. In this case,

we have chosen to incorporate P into a Ni site to investigate how a local P binding

intimately to other metals through an oxygen may affect the energetics. We note if our

6

experimental catalyst was a crystalline LDH, that a γ-phase NiOx structure with PO43-

intercalating between the layers would be accurate.

To be thorough, we have performed additional calculations where the P is

incorporated as a PO43- molecule intercalated into the structure of the NiOx catalysts

(Supplementary Fig. 24). It was found that when PO43- is intercalated, the ΔGoxidation

was higher than when the P replaces a Ni in the structure. Interestingly, the doping with

different elements influenced the energies far less when intercalated in between layers

than when substituted. For example, ΔG(Ni2+ Ni3+) is lowered by 0.51 eV when the

P-substituted NiOxP is doped to NiFeCoOxP, but the ΔG(Ni2+ Ni3+) is only lowered by

0.11 eV when NiOx-PO4 is doped to NiFeCoOx-PO4. Furthermore, the ΔG(Ni3+ Ni4+) is

not lowered by any significant amount when phosphate is intercalated between the

layers suggesting that the Ni4+ phase is less likely with phosphate between the layers

than when P is substituted into a Ni site. We performed P K-edge of FePO4 and

NiCoFeP at Beijing Light Source 4B7A beamline to illustrate the structure

(Supplementary Fig. 5). We confirmed that P incorporated into NiCoFeP as phosphate

form. We also note that our ICP-AES elemental analysis showed no leaching of the

constituents. If P was incorporated as a phosphate in between layers of the metal

oxyhydroxide catalyst, then presumably it would be easy for this phosphate to leach

into solution. Therefore, we believe that P is intimately incorporated into the

amorphous catalyst as a phosphate and not in between layers and our computational

model is a reasonable approximation of our system. Therefore, we believe our models

are reasonable considering that the experimental spectroscopy data gives evidence for

Ni4+ only upon doping with Fe and Co which is in agreement with our simulations.

To test possible doping positions, we have randomly changed the positions of Fe,

Co, and P in the NiO2 compound. It was found that the formation energies for all

compounds were within 0.03 eV of each other. Furthermore, the lattice constants

between all configurations were found to be almost identical. Therefore, we believe

that the specific doping positions would not significantly influence the formation

energies or structures. The structures, energies, and lattice constants are shown below

in Supplementary Fig. 22 and Supplementary Table 4.

7

To account for the thermodynamics of the electrochemical environment we have

calculated the ΔG(Ni2+Ni3+) and ΔG(Ni3+ Ni4+) as a function of pH. The slope of the

line is given by E0 = ΔG°f (Ni3+) – ΔG°f (Ni2+) – 0.05916*pH for the Ni2+ to Ni3+

oxidation and E0 = ΔG°f (Ni4+) – ΔG°f (Ni3+) – 0.05916*pH for the Ni3+ to Ni4+

oxidation. The plot (Supplementary Fig. 28) shows that at pH=7 the anodic potential

window for Ni3+ to exist is much larger (0.2V – 1.1V) for the doped NiCoFeP catalyst

than for Ni alone (1.07V – 1.29 V). Notice that at the applied potentials of 1.4V and

above the Ni4+ species should be stable.

L-edge simulation with the OCEAN code

The formalism of the Bethe-Salpeter equation can be found in Ref (30). The input

DFT+U calculations are performed using norm-conserving pseudopotentials. Very

large energy cutoff of 160 Ryd is used for the plane-wave basis set of the Kohn-Sham

wavefunctions. The key part of the electron-core-hole interactions is the screened

Coulomb interaction, which is obtained within the random phase approximation. A

large number of empty bands up to 50 eV above the Fermi level are included in the

calculations. A 6x6x6 k-grid over the Brillouin zone is used to produce the

sXASspectra, which is convoluted with a 0.5 eV Lorentzian line shape.

Theoretical sXAS spectra of Ni2+ with multiplet effects is shown below

(Supplementary Fig. 3). This spectrum is produced with a crystal field splitting of 2.0

eV, i.e., the 10Dq parameter used in the CTM4XAS code, which is estimated from the

DFT+U calculation for the layered Ni(OH)2 structure.

Because the multiplet calculations cannot accurately reproduce the relative

alignment of different oxidation states, we adopt a hybrid spectral simulation scheme,

i.e., using multiplet calculations to recapture the missing satellite features and relying

on the screened Coulomb interaction used in the Bethe-Salpeter Equation to produce

the relative energy shift from Ni2+ to Ni4+. More specifically, our comparison with

experiments is carried out as follow:

(a) Obtain the Ni2+ spectrum from CTM4XAS (blue curve in Supplementary Fig.

8

3) and align it with the Ni2+ experimental spectrum (black). The magnitude of

the spin-orbit coupling and the multiplet peaks are in excellent agreement with

the experiment. The absolute energy alignment with experiment for Ni2+,

ECTM, is thus determined.

(b) In CTM4XAS, the multiplet splitting can be turned off by setting the Slater

integrals Fpd = Gpd = 0, which means there is no Coulomb interaction

between the electrons on the 2p core level and those on the 3d shell. Switching

off the multiplet effects yields a spectrum with no side satellite features (red).

Using the same energy shift (ECTM) as obtained in the last step which indicates

that the multiplet effects tend to split the spectrum into two main peaks. The

lower energy peak of this pair was the one we aligned to experiment above.

(c) We propose that the multiplet-free spectrum in (b) enables a fair comparison

between the CTM4XAS and the OCEAN calculation, which cannot describe

multiplets. We therefore align the Ni2+ BSE spectrum (orange) with the

multiplet-free one (red), which yields an absolute energy alignment for the

BSE spectral simulation, EBSE.

(d) Using EBSE and the relative energy shift from Ni2+ to Ni4+ as predicted by

BSE, we can determine the absolute alignment of the Ni4+ BSE spectrum

(green).

(e) The Ni4+ BSE spectrum now appears at a slightly higher energy, about ~ 1 eV

than the experimentally determined L3 edge of Ni4+ (gray).

9

Materials and Methods

Chemicals

Iron (III) chloride (FeCl3), nickel (II) chloride hexahydrate (NiCl2•6H2O), cobalt (II)

chloride (CoCl2), sodium dihydrogen phosphate (NaH2PO4•2H2O), ethanol (≥99.5%),

propylene oxide (≥99%), Nafion solution (5wt%) and N, N-Dimethylformamide were

purchased from Sigma-Aldrich. All the chemicals were used without further

purification.

Synthesis of gelled oxyhydroxides

Gelled NiCoFeP oxyhydroxides were synthesized using a modified aqueous sol-gel

technique19. NiCl2•6H2O (2.45 mmol), anhydrous FeCl3 (0.27 mmol), and CoCl2

(0.27 mmol) were first dissolved in ethanol (2 mL) in a vial. NaH2PO4•2H2O (0.27

mmol) was dissolved in another solution of deionized water (DI) (0.23 mL) mixed

with ethanol (2 mL) in a separate vial. All solutions mentioned above were cooled in

an ice bath for 2 h in order to prevent uncontrolled hydrolysis and condensation which

may lead to the formation of precipitate rather than gel formation. The Ni, Fe, and Co

precursors were then mixed with the supernatant of ethanol-water mixture containing

NaH2PO4•2H2O to form a clear solution. To this solution, propylene oxide (≈1 mL)

was then slowly added, forming a gel. The NiCoFeP wet-gel was aged for 24 hours to

promote network formation, immersed in acetone, which was replaced periodically

for 5 days before the gel was supercritically dried using CO2.

Gelled NiCoP oxyhydroxides were synthesized following a process similar to

that of NiCoFeP oxyhydroxides except with the addition of FeCl3 as iron precursor.

NiCl2•6H2O (2.45 mmol) and CoCl2 (0.27 mmol) were first dissolved in ethanol (2

mL) in a vial. The rest steps and parameters are the same.

Gelled NiP oxyhydroxides were synthesized following a process similar to

that of NiCoFeP oxyhydroxides except with the addition of FeCl3 and CoCl2 as iron

and cobalt precursors. NiCl2•6H2O (2.7 mmol) were first dissolved in ethanol (2 mL)

in a vial. The rest steps and parameters are the same.

10

All other gelled controls (NiCoFe, NiCo and Ni oxyhydroxides) with no

phosphorus were synthesized following a process similar to those with phosphorus,

except with the addition of NaH2PO4•2H2O in the mixture of water and ethanol.

Characterization

Transmission electron microscopy (TEM), high resolution TEM images (HRTEM),

high-angle annular dark-field scanning transmission electron microscopy

(HAADF-STEM), and electron energy loss spectroscopy (EELS) elemental maps

after OER were taken on a Hitachi 2700C transmission electron microscope operated

at 200 kV. The samples were prepared by dropping catalyst powder dispersed in

ethanol onto carbon-coated copper TEM grids (Ted Pella, Redding, CA) using

micropipettes and were dried under ambient conditions. The cross-section scanning

electron microscopy images of thin film samples were acquired using a Hitachi

SU8230 scanning electron microscope operated at 1.0 kV. Powder X-ray diffraction

(XRD) patterns were obtained with MiniFlex600 instrument. Data were collected in

Bragg-Brettano mode using 0.02° divergence with a scan rate of 0.1° s-1. The atomic

composition of samples was determined using Inductively Coupled Plasma Optical

Emission Spectrometer (ICP-AES). Mass specific surface areas of the catalysts were

determined via nitrogen Brunauer, Emmett, and Teller (BET) area measurements. N2

adsorption-desorption isotherms of the catalysts were measured at -196oC using

Quantachrome Autosorb-1 MP analyzer. Prior to the measurements, the samples were

outgassed under vacuum for 10 h at 100°C. The total surface areas are deduced from

the isotherm analysis in the relative pressure range from 0.05 to 0.20. The masses of

the samples were measured with an ultramicro balance.

In-situ X-ray absorption

X-ray absorption measurements at the nickel, Co and iron L-edges were performed at

the spherical grating monochromator (SGM) beamline 11ID-1 at the Canadian Light

Source. The window of the sample cells was mounted at an angle of roughly 45° with

respect to both the incident beam and the detectors. All measurements were made at

11

room temperature in the fluorescence mode using Amptek silicon drift detectors

(SDDs) with 1024 emission channels (energy resolution ∼120 eV). Four SDDs were

employed simultaneously. The scanning energy ranges of Ni, Co and Fe L-edges were

set between 840 and 900 eV, 760 and 820 eV, and 690 and 750 eV, respectively. For

every edge, the scanning time was 30 s and repeated for ten times, and the

fluorescence of every edge was collected at the same absorption edge. The partial

fluorescence yield (PFY) was extracted from all SDDs by summation of the

corresponding metal L emission lines.

Saturation effects in the PFY measurements were noted in the ex-situ

measurement due to the relatively large intensity of the multiplet feature in

comparison with the main peak. While some saturation effects may be present in the

data, the interpretation of the in-situ results are not consistent with simple saturation.

In particular, the appearance of the high energy resonance on the L2 peak, which

appears upon application of the bias voltage, could not result from spectral distortions.

The probing depth of PFY was evaluated via the composition and the density,

assuming that the valence of Ni, Co and Fe in hydroxides are 2+, 2+ and 3+

respectively. The densities of Ni(OH)2, Co(OH)2 and Fe(OH)3 are set as 4.1 g cm-3,

3.6 g cm-3, and 3.9 g cm-3. And the densities of our samples were calculated via the

molar ratio of different unitary hydroxides. The estimated probing depth is around

300 nm. Even though the thickness of our sample is larger than the probing depth

(Supplementary Fig. 6), the nanoporous structure makes thick film possible for in situ

SXAS.

NiO (Ni2+), LiNiO2 (Ni3+), and potassium nickel(IV) paraperiodate

K2Ni(H2IO6)2 (Ni4+) references were measured at Beijing Light Source using the

4B9B beamline (Supplementary Fig. 4). K2Ni(H2IO6)2 (potassium nickel(IV)

paraperiodate) is known to be prone to reduction. The blue curve is a mixture of Ni2+

and Ni4+ since Ni4+ is unstable (Supplementary Fig. 4a). The second peak of Ni

L3-edge at 857.5 eV and Ni L2-edge around 875.1 eV decreased after exposing to air,

12

indicating that these two peaks are the characteristic peak of Ni4+ (Supplementary Fig.

4b). To obtain the spectrum for Ni4+, we subtracted Ni2+ spectrum from the spectrum

of potassium nickel(IV) paraperiodate K2Ni(H2IO6)2 (Ni4+) reference and obtained a

single-peak shape that agrees well with the simulated Ni4+ spectra (Fig. 2c). The

L-edge spectra were calibrated by using the total electron yield spectrum of NiO,

which have absorption maxima at 853.2 eV29. Calibration spectra were run before in

situ experiment to monitor beam stability and resolution. Considering LiNiO2

reference is air sensitive, we stored and prepared LiNiO2 in a glove box.

To better clarify the advantage of soft x-ray absorption, we performed Ni

K-edge of NiP, NiCoP, NiCoFeP at 1.8 V vs. RHE, and compared with that of Ni

L-edge (Supplementary Fig. 6). Even though hard XAS can reveal d-states

(Supplementary Fig. 6a), more details of 3d electronic structures have been blurred

due to the 1s lifetime broadening and the forbidden dipole 1s-3d excitation, whereas

soft XAS probes directly the 3d-states at the valence band and provides information

about the electronic structure around the excited atom, which is sensitive to the local

structure and chemical environment of 3d metals (Supplementary Fig. 6b). Ni K-edge

could not give the accurate ratio of different oxidation states.

Sample Cell for in-situ X-ray absorption

The bodies of the sample cells were fabricated on an Object Connex500 printer by 3D

printing with DurusWhite material. The design of the electrochemical flow cell was

adapted from a previously used flow cell46. Silicon nitride membrane windows (1 mm

× 1mm × 100 nm) in Si frames (5 mm × 5mm × 525 μ m) were purchased from SPI

Supplies. For electrochemical flow cells, the windows were treated by HF, and then

coated by electron-beam evaporated titanium (10 nm) and gold (30 nm). The catalysts

were coated on the windows via drop-casting with a thickness of 400 nm, which was

shown in fig. S5 below prepared via the same procedure and parameters on FTO. The

gold-covered window was contacted by gold wires. Silver and platinum wires were

used as reference and counter electrodes. In the final step of assembly, the windows

were glued to the cell body using Varian Torr seal epoxy, forming an electrolyte

13

chamber of ∼1 mm height. The contact between the gold film on the windows of the

electrochemical cell and the gold wire was monitored during the curing process of the

epoxy.

Electrochemical Measurements

Electrochemical measurements were performed using a three-electrode system

connected to an electrochemical workstation (Autolab PGSTAT302N) with built-in

electrochemical impedance spectroscopy (EIS) analyzer. The working electrode were

Au (111) single crystal electrodes purchased from MTI Corporation (U.S.). Ag/AgCl

(with saturated KCl as the filling solution) and platinum foil were used as reference

and counter electrodes, respectively.

To load the thin film of catalysts on Au (111), 4 mg of catalyst was dispersed

in 4 ml N, N-Dimethylformamide. The thin layer was pin coated using the following

parameters: spin speed = 5000 revolutions per minute (rpm), acceleration = 2500 rpm

s−1 and spin time = 30 s.

To load the catalyst on a gold-plated Ni foam (thickness: 1.6 mm, Sigma), 20

mg of catalyst was dispersed in a mixture containing 2 ml of water and 2 ml ethanol,

followed by the addition of 100 μL Nafion solution. The suspension was sonicated for

30 min to prepare a homogeneous ink. Gold-plated Ni foam with a fixed area of 0.5 x

0.5 cm2 coated with water resistant silicone glue was drop-casted with 20 μL of the

catalyst ink.

Cyclic voltammetry (CV) measurements at 50 mV s-1 were performed for 3

cycles prior to the record of linear scan voltammetry (LSV) at 1 mV s-1 for each

sample. A diluted paste was prepared by dispersing 1 mg of catalyst powder in 1 ml N,

N-Dimethylformamide, sonicating for 30 mins. Then, the working electrode was

prepared by dripping 5 μl catalyst ink onto GCE.

The integrating reducing CO2 into CO and oxidizing H2O to O2 were

performed in a gas-tight two-compartment H-cell separated by a bipolar membrane

(Fumasep FBM, purchased from Fuel Cell Store). CO2 gas was delivered into the

cathodic compartment at a rate of 5.00 sccm and was routed into gas chromatograph

(GC, PerkinElmer Clarus 600). The GC was equipped with a Molecular Sieve 5A

14

capillary column and a packed Carboxen-1000 column. Argon (Linde, 99.999%) was

used as the carrier gas. The GC columns led directly to a thermal conductivity

detector (TCD) to quantify hydrogen and a flame ionization detector (FID) equipped

with a methanizer to quantify carbon monoxide.

The number of active sites can be calculated via the following equation (3):

CoNi

ACo

Ni

CoNi

rF

NQn

(3)

where QNi is the integration area of Co redox peak from CV curves, NA is

Avogadro's constant, F is the Faraday constant, rNi/Co is the molar ratio of Ni/Co,

assuming that Ni2+/Ni4+ is a two-electron process, Co2+/Co3+ is a one-electron process.

The TOF value was calculated from equation (4):

nF

AJTOF

**4

** (4)

J is obtained at overpotential = 350 mV, A is the geometric area, F is the Faraday

constant and η is the Faradaic efficiency. n is the mole number of active atoms on the

electrode, calculated from equation (3) above.

The parameters of NiCoFeP, NiCoP, and NiP are listed as Supplementary Table 5.

Arrhenius Apparent Activation Energy Determination

The sealed three-electrode glass cell was suspended in a thermostatic silicone

oil bath. The electrodes were passed through holes in the lid with an O-ring slightly

larger than the electrodes to minimize evaporative losses. Two holes for let-in and

let-out tubes with CO2 gas were also sealed with O-rings. The effect of temperature

on the performance of the catalysts is clearly visible in Supplementary Fig. 13. As

expected, the kinetics of the OER are increased at elevated temperatures, reflecting

the temperature dependence of the chemical rate constant. The apparent

electrochemical activation energy (Ea) for water oxidation can be determined using

the Arrhenius relationship46

k(log )

(1/ ) 2.3

ai E

T R

(5)

15

where ik is the kinetic current at η = 700 mV, T is the temperature, and R is the

universal gas constant. The Arrhenius plots at η = 700 mV for NiCoFeP catalysts and

controls are shown in Fig. 3c and Supplementary Fig. 14, which reveal a linear

dependence on temperature. From the slope of the Arrhenius plot, the Ea can be

extracted, listed in Table 1. NiCoFeP has a lowest activation energy of 27.00 kJ mol-1.

Calibration of the reference electrode Ag/AgCl and Ag wire

The calibrations of Ag/AgCl reference electrode and Ag wire reference electrode

were conducted in the standard three-electrode system (the same system as that for

performance measurements) as reference electrodes, using Pt foil as working and

counter electrodes. The 0.5M H2SO4 electrolyte (pH = 0.3) was pre-purged with

Grade 4 H2 overnight and continuously bubbled with H2 with a flow rate of 25 ml

min-1 during calibrations. LSV was run around +/- 100 mV between hydrogen

evolution and oxidation, and the potential of zero current was recorded. The LSV

curves were shown in Supplementary Fig. 16 below. The potential of zero current was

around 0.2198V (including pH correction of 0.3*0.0591=0.0177V) for Ag/AgCl

electrode and 0.3306V (including pH correction of 0.3*0.0591=0.0177V) for Ag wire

electrode, i.e. EAg/AgCl(KCl, sat) + 0.1108V = EAg wire, which is the same as the previous

reported value48. Unless otherwise stated, all experiments were performed at ambient

temperature (22 ± 2°C) and electrode potentials were converted to the RHE scale

using E(RHE) = E(Ag/AgCl) + 0.202V + 0.059*pH and E(RHE) = E (Ag wire) +

0.313V + 0.059*pH

16

Supplementary Figures and Tables

Supplementary Figure 1. DFT calculated total density of states for NiO2 (black)

and metal doped NiCoFeO2 (red). Doping with Fe and Co introduces more

electronic states between -0.5 eV and 1 eV, which may contribute to intermediate

binding and increased catalytic activity.

17

Supplementary Figure 2. Projected density of states of NiCoFeO2 which has been

extracted from the total DOS. The d-state contributions of Ni (red), Fe (blue), and

Co (O) are shown. A zoom in graph between -1 eV and 1 eV show that Ni and Fe

both have peaks in the pDOS whereas Co is essentially zero. This suggests that Ni and

Fe can both contribute to catalytic activity whereas Co is inactive.

18

Supplementary Figure 3. Theoretical sXAS at the Ni L-edge of Ni2+ and Ni4+ by

CTM4XAS and BSE calculations.

19

Supplementary Figure 4. (a) Ni L-edge of potassium nickel(IV) paraperiodat

K2Ni(H2IO6)2 (Ni4+) reference test immediately (black), and exposed to air for 5 hours

(red). (b) Ni L-edge of NiO (Ni2+), LiNiO2 (Ni3+), and potassium nickel(IV)

paraperiodat K2Ni(H2IO6)2 (Ni4+) references after subtracting Ni2+ signals.

20

Supplementary Figure 5. P K-edge of FePO4 reference and NiCoFeP catalysts.

21

Supplementary Figure 6. Ni K-edge (a) and L-edge (b) x-ray absorption of NiP,

NiCoP, and NiCoFeP at 1.8 V vs. RHE, respectively.

22

Supplementary Figure 7. Characterization of NiCoFeP oxyhydroxides catalysts.

a, High-angle-annular-dark-field (HAADF) STEM image of nanoporous structure of

NiCoFeP. b, Selected area electron diffraction (SAED) pattern. c, Energy-dispersive

X-ray spectroscopy (EDS) elemental mapping from NiCoFeP oxyhydroxides sample.

23

Supplementary Figure 8. X-ray diffraction (XRD) of NiCoFeP oxyhydroxide.

NiCoFeP oxyhydroxide catalyst revealed no evidence for a crystalline phase.

24

Ni Fe

Co O

Supplementary Figure 9. Electron energy loss spectroscopy (EELS) elemental

mapping from NiCoFeP oxyhydroxides catalysts.

25

Supplementary Figure 10. Calibration curve of inductively coupled plasma

atomic emission spectroscopy (ICP-AES) and ICP-AES spectra of NiCoFeP

oxyhydroxides: Ni element (a, b), Co element (c, d), Fe element (e, f) and P

element (g, h).

26

Supplementary Figure 11. Cross-section SEM image of NiCoFeP oxyhydroxides

on FTO.

27

Supplementary Figure 12. Linear combination analysis of in situ Ni L-edge

X-ray absorption spectra of NiCoFeP catalysts for Ni4+, Ni3+, and Ni2+ at 0V (a),

1.6V (b), at 1.8V (c) and at 2.0V (vs. RHE) (d). The black curves are the original

data, the red curves are the fitting results, the blue filling areas are pure Ni2+ spectra,

the cyan filling areas are pure Ni4+ spectra and the pink filling areas are pure Ni4+

spectra.

28

Supplementary Figure 13. Linear combination analysis of in situ Ni L-edge

X-ray absorption spectra of NiCoP catalysts for Ni4+, Ni3+, and Ni2+ at 1.6V (a),

at 1.8V (b) and at 2.0V (vs. RHE) (c). The black curves are the original data, the red

curves are the fitting results, the blue filling areas are pure Ni2+ spectra, the cyan

filling areas are pure Ni4+ spectra and the pink filling areas are pure Ni4+ spectra.

29

Supplementary Figure 14. Linear combination analysis of in-situ Ni L-edge

X-ray absorption spectra of NiP catalysts for Ni4+, Ni3+, and Ni2+ at 1.6V (a), at

1.8V (b) and at 2.0V (vs. RHE) (c). The black curves are the original data, the red

curves are the fitting results, the blue filling areas are pure Ni2+ spectra, the cyan

filling areas are pure Ni4+ spectra and the pink filling areas are pure Ni4+ spectra.

30

Supplementary Figure 15. Fe L-edge X-ray absorption spectra of NiCoFeP

catalyst at open circuit and at +1.6V (vs. RHE).

31

Supplementary Figure 16. Co L-edge X-ray absorption spectra of NiCoFeP (a) and

NiCoP (b) catalysts at open circuit and at applied potentials (vs. RHE).

32

Supplementary Figure 17. N2 sorption isotherms of (a) NiP, (b) NiCoP and (c)

NiCoFeP oxyhydroxide; (d) LSV curves of NiCoFeP oxyhydroxides catalysts and

controls in three configurations in CO2 saturated 0.5 M KHCO3 aqueous electrolyte

loaded on Au foam and normalized by BET surface area.

33

Supplementary Figure 18. Electrochemical impedance spectroscopy (EIS) data for

NiCoFeP oxy/hydroxides catalysts and controls in three-electrode configuration in

CO2-saturated 0.5 M KHCO3 aqueous electrolyte. The data were collected for the

electrodes under 1.0 V vs. Ag/AgCl.

34

Supplementary Figure 19. The OER polarization curve of NiCoFeP

oxyhydroxides catalysts and controls in three-electrode configuration in

CO2-saturated 0.5 M KHCO3 aqueous electrolyte loaded on Au (111) with scan

rate 1 mV s-1 at 18oC, 25oC, 30oC and 35oC, respectively. We considered the

temperature effect when converting applied potential to relative hydrogen electrode

potential.

35

Supplementary Figure 20. Linear scanning voltammetric curves, sweeping the

potential about +/-100 mV between hydrogen evolution reaction and hydrogen

oxidation reaction, in three-electrode configuration in 0.5 M H2SO4 aqueous

electrolyte bubbled with pure hydrogen gas, with scan rate 0.5 mV s-1, at room

temperature.

a

b

36

Supplementary Figure 21. Linear scanning voltammetric curves of NiCoFeP

oxyhydroxides (right) and Au needles as CO2 redcution catalysts (left).

Theoretical oxidation and reduction potentials are E0O2/H2O= 1.23 V vs. RHE and

E0CO2/CO= -0.11 V vs. RHE, respectively. Dashed lines represent the overpotential

of OER and CO2 reduction at 10 mA/cm2.

37

Supplementary Figure 22. Linear scanning voltammetric curves of NiCoFeP

oxyhydroxides catalysts and Au needles in two-electrode configuration in

CO2-saturated 0.5 M KHCO3 aqueous electrolyte with scan rate 5 mV s-1.

38

Supplementary Figure 23. Chronopotentiometric curves of NiCoFeP

oxyhydroxides catalysts and Au needles in two-electrode system with constant

current densities of 10 mA cm-2, and the corresponding Faradaic efficiency from

gas chromatography measurement of evolved CO.

39

Supplementary Figure 24. (a) scheme of NiCoFeP-Au needles electrolysis; (b)

scheme of FeCoW-Au needles electrolysis with bipolar membrane.

40

Supplementary Figure 25. Structures of NiFeCoO2P with differently doping

positions. Ni (green), Fe (brown), Co (pink), P (orange), O (red).

41

Supplementary Figure 26. Ball and stick representation of the optimal Ni4+

catalyst NiFeCoO2P with top view (left), front view (middle), and side view (right)

where nickel is green, oxygen is red, cobalt is pink, iron is blue, and phosphorus

is orange.

42

Supplementary Figure 27. ΔG of Ni2+ Ni3+ and Ni3+ Ni4+ where phosphate is

intercalated between the layers of NiOx (red), NiFeOx (blue), NiCoOx (green),

NiFeCoOx (purple).

43

Supplementary Figure 28. A plot of calcualted ΔG(oxidation) as a function of pH for

pure Ni (red) and NiFeCoP(blue).

44

Supplementary Figure 29. Ni L-edge of NiCoFeP under an applied potential of 1.8

V vs. RHE fitted by (a) a combination of NiO (Ni2+), LiNiO2 (Ni3+) references or (b) a

combination of NiO (Ni2+), LiNiO2 (Ni3+), and potassium nickel(IV) paraperiodate

K2Ni(H2IO6)2 (Ni4+) references, respectively. The black curves are the original data,

the red curves are the fitting results, the blue filling areas are pure Ni2+ spectra, the

cyan filling areas are pure Ni4+ spectra and the pink filling areas are pure Ni4+ spectra.

45

Supplementary Table 1. Lattice constants for all DFT optimized structures.

a b c α β γ

Ni(OH)2 3.14

(3.13)a

3.21

(3.13)a

4.66

(4.61)a

90.00

(90.00)a

89.96

(90.00)a

120.82

(120.00)a

β-NiOOH 2.97

(2.93)b

2.89

(2.96)b

4.95

(4.84)b

82.66

(80.00)b

93.25

(90.00)b

119.96

(119.70)b

NiO2 2.76 2.76 5.08 90.07 89.60 120.03

Ni(OH)2P 3.17 3.24 4.71 89.72 90.14 121.53

β-NiOOHP 2.98 2.89 4.94 82.96 94.54 119.78

NiO2P 2.76 2.76 5.10 90.15 89.60 120.03

NiCo(OH)2 3.14 3.22 4.68 90.13 89.75 120.90

β-NiCoOOH 2.96 2.91 4.94 83.77 92.11 120.26

NiCoO2 2.77 2.78 5.12 90.52 89.44 120.04

NiCo(OH)2P 3.13 3.24 4.70 90.27 89.74 120.36

β-NiCoOOHP 2.96 2.91 4.90 83.41 92.00 119.97

NiCoO2P 2.77 2.77 5.17 91.01 89.37 119.99

NiCoFe (OH)2 3.16 3.23 4.68 90.15 89.79 120.91

β-NiCoFeOOH 2.98 2.93 4.96 83.98 91.85 120.42

NiCoFeO2 2.78 2.79 5.09 90.09 59.59 120.12

NiCoFe (OH)2P 3.14 3.24 4.69 90.31 89.94 120.15

β-NiCoFeOOHP 2.97 2.92 4.92 83.24 92.82 119.98

NiCoFeO2P 2.79 2.78 5.17 90.43 89.36 120.08

Numbers listed in brackets are from literature. a(Solid State Ionics 2010, 181, 1764− 1770),

b(J. Phys. Chem. C 2015, 119, 24315−24322).

46

Supplementary Table 2. DFT energies and Gibbs Formation energies for all

structures.

Energy (eV/fu) ΔG°f

Ni(OH)2 -24.89 -2.88

NiOOH -19.68 -1.39

NiO2 -14.25 0.31

Ni(OH)2P -24.86 -2.85

NiOOHP -19.98 -1.70

NiO2P -14.56 -0.01

NiCo(OH)2 -25.06 -2.93

NiCoOOH -20.22 -1.82

NiCoO2 -14.95 -0.28

NiCo(OH)2P -24.98 -2.86

NiCoOOHP -20.50 -2.11

NiCoO2P -15.24 -0.57

NiCoFe (OH)2 -25.33 -3.01

NiCoFeOOH -20.52 -1.93

NiCoFeO2 -15.21 -0.36

NiCoFe (OH)2P -25.22 -2.91

NiCoFeOOHP -20.85 -2.27

NiCoFeO2P -15.59 -0.74

47

Supplementary Table 3. Comparison of CO2 to CO system efficiency.

CO2 to

CO

Membrane OER catalysts

CO2 reduction

catalysts

Overpotential

at 10 mA/cm2

efficiency Reference

1 Membrane-free NiCoFeP (neutral)

Au needles

(neutral)49

1.99 V 64% This work

2 Bipolar membrane FeCoW (alkaline)16

Au needles

(neutral)49

2.37 V 54% This work

3 Nafion membrane Co-Pi (neutral)50

Iron porphyrin

(neutral)

2.5 V# 52%# Ref (51)

4 Membrane-free IrO2 (neutral) Au (neutral) 2.3 V 49% Ref (3)

5 Nafion membrane

Perfluorinated

cobalt (neutral)

Perfluorinated

cobalt (neutral)

3.2 V 41.9% Ref (52)

6 Nafion membrane

Ru complex

(neutral)

Ru complex

(neutral)

3.4 V# 39.4%# Ref (53)

# Cell potential at 1 mA/cm2.

48

Supplementary Table 4. Formation energies and lattice constants for all

NiCoFeO2P configurations.

Configuration Eform (eV) a b c α β γ

1 -0.74 2.79 2.78 5.17 90.43 89.36 120.08

2 -0.75 2.78 2.78 5.12 90.35 89.66 120.05

3 -0.72 2.78 2.78 5.16 90.30 89.43 120.01

4 -0.74 2.78 2.79 5.11 90.16 89.66 120.16

5 -0.75 2.78 2.78 5.12 90.13 89.60 120.10

6 -0.74 2.78 2.78 5.13 90.20 89.68 119.93

49

Supplementary Table 5. Comparison of catalytic parameters of NiCoFeP, NiCoP,

and NiP.

Samples

Overpotential

(mV) at 10

mA/cm2

Tafel slope

(mV dec-1)

TOF calculated by

intergration of

Ni/Co redox

BET surface area

(m2/g)

ΔH (kJ mol-1) at

η=700 mV

Rs from EIS (Ω)

NiP 588 98 0.26 295.6 54.14 43

NiCoP 547 85 0.48 129.7 37.36 42

NiCoFeP 330 60 0.68 185.4 27.00 42

50

Structures of Models Used

Structural models and the VASP POSCAR files of all optimized structures used in this study

are listed in a separate file.

51

References and notes

36 Kazimirov, V. Y. et al. Atomic structure and lattice dynamics of Ni and Mg

hydroxides. Solid State Ionics 181, 1764-1770, (2010).

37 Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector

augmented-wave method. Phys. Rev. B 59, 1758-1775 (1999).

38 Solcà, R. et al. Efficient implementation of quantum materials simulations on

distributed CPU-GPU systems. Phys. Rev. B 54, 1-12, (2015).

39 Li, Y.-F. & Selloni, A. Mechanism and Activity of Water Oxidation on

Selected Surfaces of Pure and Fe-Doped NiOx. ACS Catal. 4, 1148-1153,

(2014).

40 Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 50,

17953-17979, (1994).

41 Miwa, K., Ohba, N., Towata, S.-i., Nakamori, Y. & Orimo, S.-i.

First-principles study on lithium borohydrideLiBH4. Phys. Rev. B 69, (2004).

42 Cococcioni, M. & de Gironcoli, S. Linear response approach to the calculation

of the effective interaction parameters in theLDA+Umethod. Phys. Rev. B 71,

(2005).

43 Costanzo F. Effect of doping beta-NiOOH with Co on the catalytic oxidation

of water: DFT+U calculations. Phys Chem Chem Phys 2016, 18(10):

7490-7501.

44 Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations.

Phys. Rev. B 13, 5188-5192, (1976).

45 Balsara, N. P. & Newman, J. Comparing the Energy Content of Batteries,

Fuels, and Materials. J. Chem. Educ. 90, 446-452, (2013).

46 Risch, M. et al. Reversibility of Ferri-/Ferrocyanide Redox during Operando

Soft X-ray Spectroscopy. J. Phys. Chem. C 119, 18903-18910, (2015).

47 Anderson, A. B. et al. Activation energies for oxygen reduction on platinum

alloys: Theory and experiment. J. Phys. Chem. B 109, 1198-1203 (2005).

52

48 Riskin, M., Basnar, B., Katz, E. & Willner, I. Cyclic control of the surface

properties of a monolayer-functionalized electrode by the electrochemical

generation of Hg nanoclusters. Chemistry 12, 8549-8557, (2006).

49 Liu, M. et al. Enhanced CO2 Reduction Catalysts via Field-Induced Reagent

Concentration. Nature accepted.

50 Liu, C. et al. Water splitting–biosynthetic system with CO2 reduction

efficiencies exceeding photosynthesis. Science 352, 1210-1213 (2016).

51 Tatin, A., Comminges, C., Kokoh, B., et al. Efficient electrolyzer for CO2

splitting in neutral water using earth-abundant materials. PNAS 113,

5526-5529 (2016).

52 Morlanés, N., Takanabe, K. & Rodionov, V. Simultaneous Reduction of CO2

and Splitting of H2O by a Single Immobilized Cobalt Phthalocyanine

Electrocatalyst. ACS Catalysis 6, 3092-3095, (2016).

53 Chen, Z. et al. Splitting CO2 into CO and O2 by a single catalyst. PNAS 109,

15606-15611, (2012).